Core Responsibilities:

• Lead end-to-end ML solution design for centrifuge operational optimization.
• Own architecture, data pipelines, model deployment on Edge, and integration with industrial systems.
• Collaborate with cross-functional teams including Dianomic Systems and ADM stakeholders.

Experience in this company:

• MCP Host & MCP Server for FogLAMP
  • Designed and implemented an LLM-powered MCP Host and FogLAMP MCP server supporting APIs, DB connections, agent persona orchestration, and dynamic tool routing between Excel, FogLAMP, and cloud services.
  • Enables enterprise and agentic workflows for edge and cloud AI systems.
  • Tags: MCP, MCP Host, API Design, FogLAMP, Edge AI, Agentic Workflows, Registry
• Excel FogLAMP Datalink
  • Developed an Excel add-in to fetch FogLAMP data, enabling analysts and operators to explore and visualize edge data directly inside Excel templates.
  • Tags: Excel Add-in, Office.js, FogLAMP, Enterprise Integration, Data Visualization
• Centrifuge Breakover Forecasting (EDGE Deployment)
  • Designed, trained, and deployed a predictive ML solution on EDGE devices to optimize centrifuge operation and reduce downtime.
  • Integrated with PLC and HMI for real-time operator feedback and decision support.
  • Tags: Stacked-LSTM, PLC, HMI, FogLAMP, Edge Compute, Postgres, Grafana, GCP
• Solids Buildup Detection (EDGE Deployment)
  • Developed a time–frequency domain signal processing and autoencoder-based solution for predictive maintenance of centrifuges.
  • Deployed on FogLAMP to monitor solids accumulation using high-frequency vibration signals from multiple sensors (8 kHz streams).
  • Tags: Autoencoder, Fourier Analysis Networks, Signal Processing, High-Velocity Data, FogLAMP, Postgres, Grafana, Edge AI
• Multi-Reference Anomaly Detector
  • Researched and planned a multi-reference anomaly detection system leveraging contrastive learning and multi-instance learning for centrifuge operations.
  • Supports operator-action and machine-response anomaly detection on vibration and process data.
  • Tags: Siamese Networks, Multi-Instance Learning, Contrastive Learning, Fourier Analysis, Vibration Data, Edge AI
• MLOps Infrastructure (Cloud Agnostic)
  • Architected and developed a cloud-agnostic, edge-ready MLOps framework on GCP (partially validated on Azure and on-prem) covering ingestion, feature engineering, training, hyperparameter tuning, evaluation, monitoring, and data validation/drift checks.
  • Evaluated Postgres vs BigQuery vs InfluxDB; set up Grafana dashboards for operational and model monitoring, with support for multi-environment dev/stage/prod, backfill, health monitoring, and logging.
  • Tags: MLOps, GCP, Azure, Grafana, Postgres, BigQuery, InfluxDB, Feature Engineering, Model Training, Drift Detection
• FogLAMP Filter and Rule Development
  • Implemented multiple filters and rules for improved data handling, anomaly detection, and sensor health monitoring:
    • foglamp-filter-moving-measures
    • foglamp-filter-data-drift
    • foglamp-rule-sensordrift (vibration sensor disconnect detection)
  • Tags: FogLAMP, Edge Compute, Signal Processing, Python, Plugin Development
• Data Infrastructure and Monitoring
  • Developed system resource and syslog monitoring modules; built improved automated data transfer scripts for reliable movement of operational data.
  • Tags: System Monitoring, Syslogs, Automation, Edge Infrastructure
• Azure Data Platform & Medallion Architecture
  • Established a Delta Lake on Azure Data Lake with a Medallion Architecture using Azure Data Factory for customer analytics and ML use cases.
  • Evaluated DVC, MLFlow, and Databricks for experiment tracking and data/version management.
  • Tags: Azure Data Factory, Delta Lake, DVC, MLFlow, Databricks, Data Medallion Architecture
• Plugin Development for Testing
  • Created foglamp-south-datafaker plugin to generate configurable signal patterns for system, pipeline, and MCP orchestration testing.
  • Tags: FogLAMP, Edge Compute, Plugin Development, Test Infrastructure

Notable Contributions:

• Defined and delivered an Edge AI and MLOps blueprint for centrifuge optimisation that is reusable across assets and sites.
• Shipped multiple production-grade EDGE deployments (forecasting, anomaly detection, solids buildup) integrated with PLC/HMI and Grafana-based monitoring.
• Introduced MCP Host + FogLAMP MCP server and Excel FogLAMP Datalink, enabling operator-friendly, tool-augmented workflows for industrial data analysis.
• Established Azure Data Platform with Medallion architecture and evaluation frameworks (DVC/MLFlow/Databricks) to standardise experimentation and governance.

Core Responsibilities:

• Build data enrichment and provider matching pipeline.
• Optimize identity resolution at scale.

Experience in this company:

• National Provider Identifier (NPI) Filler services for the missing NPI
  • Filled in the missing information for multiple cases of Healthcare providers by utilising the data available from data source.
  • Tags: Identity Resolution, Similarity Mapping, Search, Parallel Processing, Big Data, APIs, Redis, Spark

Core Responsibilities:

• Scale Day Ahead Capacity Planning from pilot to multi-country production.
• Develop monitoring, automation, and utilities for faster deployment cycles.
• Work with operations, supply chain, and engineering teams across geographies.

Experience in this company:

• Day Ahead Capacity Planning
  • Providing Day Ahead forecast for the Last Mile, to the Delivery Stations. This enables them to plan for the logistics.
  • Cost Saving comes from minimizing the over- and the underbooking of the logistics.
  • Tags: Time Series, Curve Matching, ML, Sagemaker, EC2, Lambda, Event Bridge, S3, Dashboard
• Common Utilities Python Lib (v1)
  • A simple library that contains the common utilities to be used as part of multiple projects.geography
  • Credential Keeping and Retrieving: This is based on AWS Secret Manager
  • SQL Query Executor: Run SQL jobs while capturing and providing the job stats and logging them too.
  • Logger: A file and table-based logger wrapper
  • S3 Interaction: A wrapper containing functions for S3-based utility.
  • Operations: A set of certain classes & functions for Path, Time, & Math-related Operations
  • Data Evaluation & Clustering: Functions to evaluate & show performance of Regression & class-based data in much greater detail.
  • Data Analysis: Function to identify Missing data with respect to groups, plotting the data, etc was present.
  • Document Interaction: ability to connect & retrieve data from Quip
  • Tags: AWS Services, DB, SQL, Evaluation, Logging, Analysis, data pipeline
• Capacity Planning Performance Evaluation, Analysis & Dashboard
  • Developed and maintained capacity planning, evaluation, and analysis. It is robust enough to get multiple insights and even generate plans for current state, pilots, and areas for improvement.
  • Helps better evaluate multiple components of delivery at multiple countries, stations, ship methods and cycles
  • Tags: Time Series, EC2, QuickSight, S3, Excel
• Multiple Reports Creation and Data Pipelines Testing
  • Created Reports for Middle Mile Performance Evaluation and tested/adjusted a few data pipelines for reporting.

Notable Contributions:

• Took over and stabilized the project with zero active maintainers.
• Led deployment expansion to multiple countries within 6 months.
• Optimized pipeline runtime and reporting latency for faster planning.

Core Responsibilities:

• Build ML POCs and MVPs for digital transformation in agriculture and supply chain.
• Collaborate with product managers and global business stakeholders.
• Transition pilot projects to other teams after successful MVP delivery.

Experience in this company:

• ML labs v2 setup
  • Defining ways to work with ML projects and other applications having ML-related features
  • Bringing sustainable practices for accelerated development while maintaining the agility
  • Planning phase of data governance, 3rd Party Data Annotators, Portal development for data tagging, selection and versioning, data collection experiment designs and pipelines.
• Estimating Growth Indicator of Chicken (POC)
  • Estimating the weight and age of the chicken based on a video recording done using the mobile camera.
  • Tags: Agile Experiment Setup & Design, Object Detection, Image Standardizer, Estimator, Ensembling
• Defective Nugget Detection and Accounting (POC then MVP)
  • Defective Nugget Detection and Tracking on conveyor belt with output projected on display for manual removal.
  • Tags: Jetson Nano, Raspberry Pi, Ardino Uno, Edge device, Object Detection, Tracking, on-premise, real-time
• HR Chatbot (Myco) (MVP)
  • Initial product version was launched for Singapore and Australia with the capability for form filling, ticket creation, ticket checking, easy data integration from excel sheet
  • Tags: Rasa, Redis, NLU, NDM, NLG, NLP, Chatbot, excel
• Financial Commentary Generation (POC)
  • A utility for table-to-text based on template method was developed for the initial stage, b/c of the lack and too much fragmentation of data source this approach was deemed fit.
• YOLOv5-v4 training and deployment integration on AWS, suitable to multiple tailored needs.
  • Tags: Computer Vision, Sagemaker, ECR, S3, Lambda Function, API Gateway, IAM, CloudWatch
• Computer Vision Utility
  • Streamlit-based web application for parsing, labelling, and visualizing annotations, with support for Yolo prediction.
  • Ability to connect work with local, S3 and pcloud storage.
• MiApp feature involving weather data visualization using the GIS data.
• Information Extraction from Ship Invoices (POC)
  • Extracting multiple information from the information that was shared in old and current emails.
  • Tags: Text-to-Table, Analytics; Regex, Python Email to DB

Notable Contributions:

• Delivered MVPs that transitioned into internal products.
• Developed first successful internal vision pilot used for later projects.
• Groomed soft and business skills through direct stakeholder interaction.

Core Responsibilities:

• Develop ML-based bot detection frameworks.
• Build and optimize detection pipelines with focus on low FP and adaptive learning.
• Research and implement reinforcement, unsupervised, and semi-supervised solutions.

Experience in this company:

• ICLSSTA
  • Solely developed a framework, which develop and evaluate unsupervised learning solution in much depth and this from a single config file.
  • Tags: Framework, Clustering, Anomaly, Outlier, Conceptual Drift Detection, Ensembling, Support all Dimension reduction, Anomaly, and Clustering Algorithm from Sklearn, Add. Custom Algorithm
• Adaptive Action Taking (AAT)
  • Developed a generalized module named which makes use of reinforcement learning to automatically take action on incoming traffic on the web property.
  • Tags: Reinforcement Learning, Multi-Armed Bandit, Smart handling of Business Limits, Self-Adjusting
• Deep Behaviour Analysis (IDBA)
  • Partially developed a module named Intent based, which utilizes LSTM to yield encoded features and scores that are used with supervised, anomaly and clustering.
  • Tags: RNN, Auto-Encoder, Semi-Supervised Learning, Sequence Based Detection, Deep Learning
• Threshold Online Reinforcement (ThOR)
  • Helped developed a module named, which makes utilizes online learning to develop a probability score distribution over of isotonic regression for action taking.
  • Tags: Isotonic Regression, Online Learning, Self-Adjusting, Smart handling of Business Limits
• Second Level Module Integration
  • Developed and implemented end to end machine and reinforce learning based solutions using ICLSSTA-AAT, IDBA-ICLSSTA-AAT, & IDBA-ThOR
  • Tags: Semi-Supervised, Supervised, unsupervised, reinforcement learning, behaviour analysis
• Titan Batch Analyser
  • Developed and Implemented, which utilizes multiple browser signatures, behaviour based rules, regular expressions and databases to generate a suspiciousness score to take action on traffic.
  • Tags: SQL & Python Implementation, ETL, Advance SQL operations, Handling Multiple Datasets
• Rule Scripts
  • Developed and Implemented multiple rules which were based on network, device and browser fingerprint, and behaviour of the visitors. These were adequately balanced between causal and correlated, yielding negligible FP.
  • Tags: Domain knowledge, Behaviour Analysis, Rules development
• Rule Mining
  • Created a Recommendation System for the highlighting possible bad signature coming in the traffic.
  • Tags: Associate Rule Mining, Text Data Pre-processing
• Developing Dynamic Moving Collective Intelligence which can evolve while working across multiple sid.
• Other Works
  • Developed Dashboard, Visualization and Analysis, spreadsheet to understand the whole behaviour of the traffic. Partially interactive visualization sheets were also developed in python.
  • Tags: SQL and Spreadsheet/Python Implementation, Visualization, Result Sharing

Notable Contributions:

• First author and co-inventor on patented bot detection frameworks:
  • “System and method for detecting bots based on iterative clustering and feedback-driven adaptive learning techniques” — US20200099713A1
  • “System and method for detecting bots using semi-supervised deep learning techniques” — US20200099714A1
• Contributed IP and research that supported ShieldSquare’s acquisition by Radware.